Parquet is columnar, you can read each column independently, and a different compression scheme can be used for each column. You first have to read the beginnin

Parquet is columnar, you can read each column independently, and a different compression scheme can be used for each column. You first have to read the beginning of the file to get the layout though (And the end of it as well, according to a library I have been using. Don't know why) . The http server must of course support range queries.